|
|
--- |
|
|
license: apache-2.0 |
|
|
pipeline_tag: image-classification |
|
|
--- |
|
|
# PeleeNet |
|
|
|
|
|
## **Use case** : `Image classification` |
|
|
|
|
|
# Model description |
|
|
|
|
|
|
|
|
|
|
|
PeleeNet is a lightweight convolutional neural network architecture designed for **efficient real-time inference on mobile and edge devices**. It is built upon dense connectivity principles, similar to DenseNet, but optimized to significantly reduce computational cost while maintaining strong accuracy. |
|
|
|
|
|
Unlike MobileNet, PeleeNet does **not rely on width multipliers or depthwise separable convolutions**. Instead, it uses **two-way dense blocks**, bottleneck layers, and efficient stem blocks to reduce memory access cost and improve speed on practical hardware. |
|
|
|
|
|
The original paper demonstrates that PeleeNet achieves competitive accuracy compared to MobileNet while requiring **fewer parameters and lower latency**, especially on real devices. |
|
|
|
|
|
(source: https://arxiv.org/abs/1804.06882) |
|
|
|
|
|
The model is quantized to **int8** using **ONNX Runtime** and exported for efficient deployment. |
|
|
|
|
|
## Network information |
|
|
|
|
|
|
|
|
| Network Information | Value | |
|
|
|--------------------|-------| |
|
|
| Framework | Torch | |
|
|
| MParams | ~2.69 M | |
|
|
| Quantization | Int8 | |
|
|
| Provenance | https://github.com/Robert-JunWang/PeleeNet | |
|
|
| Paper | https://arxiv.org/abs/1804.06882 | |
|
|
|
|
|
## Network inputs / outputs |
|
|
|
|
|
|
|
|
For an image resolution of NxM and P classes |
|
|
|
|
|
| Input Shape | Description | |
|
|
| ----- | ----------- | |
|
|
| (1, N, M, 3) | Single NxM RGB image with UINT8 values between 0 and 255 | |
|
|
|
|
|
| Output Shape | Description | |
|
|
| ----- | ----------- | |
|
|
| (1, P) | Per-class confidence for P classes in FLOAT32| |
|
|
|
|
|
|
|
|
## Recommended platforms |
|
|
|
|
|
|
|
|
| Platform | Supported | Recommended | |
|
|
|----------|-----------|-----------| |
|
|
| STM32L0 |[]|[]| |
|
|
| STM32L4 |[]|[]| |
|
|
| STM32U5 |[]|[]| |
|
|
| STM32H7 |[]|[]| |
|
|
| STM32MP1 |[]|[]| |
|
|
| STM32MP2 |[]|[]| |
|
|
| STM32N6 |[x]|[x]| |
|
|
|
|
|
# Performances |
|
|
|
|
|
## Metrics |
|
|
|
|
|
- Measures are done with default STM32Edge.AI configuration with enabled input / output allocated option. |
|
|
- All the models are trained from scratch on Imagenet dataset |
|
|
|
|
|
### Reference **NPU** memory footprint on Imagenet dataset (see Accuracy for details on dataset) |
|
|
|
|
|
| Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STEdgeAI Core version | |
|
|
|-------|---------|--------|------------|--------|--------------|--------------|---------------|----------------------| |
|
|
| [peleenet_pt_224](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/peleenet_pt/Public_pretrainedmodel_public_dataset/Imagenet/peleenet_pt_224/peleenet_pt_224_qdq_int8.onnx) | Imagenet | Int8 | 224×224×3 | STM32N6 | 1421 | 0 | 2751.06 | 3.0.0 | |
|
|
|
|
|
|
|
|
### Reference **NPU** inference time on Imagenet dataset (see Accuracy for details on dataset) |
|
|
|
|
|
|
|
|
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version | |
|
|
|--------|----------|--------|-------------|------------------|------------------|---------------------|-----------|-------------------------| |
|
|
| [peleenet_pt_224](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/peleenet_pt/Public_pretrainedmodel_public_dataset/Imagenet/peleenet_pt_224/peleenet_pt_224_qdq_int8.onnx) | Imagenet | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 18.01 | 55.52 | 3.0.0 | |
|
|
|
|
|
|
|
|
### Accuracy with Imagenet dataset |
|
|
|
|
|
Dataset details: [link](https://www.image-net.org) |
|
|
Number of classes: 1000. |
|
|
To perform the quantization, we calibrated the activations with a random subset of the training set. |
|
|
For the sake of simplicity, the accuracy reported here was estimated on the 50000 labelled images of the validation set. |
|
|
|
|
|
|model | Format | Resolution | Top 1 Accuracy | |
|
|
|---------|--------|------------|----------------| |
|
|
| [peleenet_pt_224](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/peleenet_pt/Public_pretrainedmodel_public_dataset/Imagenet/peleenet_pt_224/peleenet_pt_224_qdq_int8.onnx) | Int8 | 224x224x3 | 70.37% | |
|
|
| [peleenet_pt_224](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/peleenet_pt/Public_pretrainedmodel_public_dataset/Imagenet/peleenet_pt_224/peleenet_pt_224.onnx)| Float | 224x224x3 | 70.57% | |
|
|
|
|
|
## Retraining and Integration in a simple example: |
|
|
|
|
|
Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services) |
|
|
|
|
|
|
|
|
|
|
|
# References |
|
|
|
|
|
<a id="1">[1]</a> - **Dataset**: Imagenet (ILSVRC 2012) — https://www.image-net.org/ |
|
|
|
|
|
<a id="2">[2]</a> - **Model**: PeleeNet — https://github.com/Robert-JunWang/PeleeNet |
